1

Vowel Reduction*

Katherine Crosswhite, University of Rochester

1. Background on Vowel Reduction

Vowel reduction is a well-known phonological phenomenon: The idea that certain

vowels might undergo qualitative changes in unstressed positions is likely familiar to anyone

who has taken an introductory phonology course. Because this phenomenon can be so succinctly

described—i.e., "unstressed vowels undergo neutralization"—it is often assumed that vowel

reduction is a unitary phenomenon, with a single formal analysis. In this chapter, I take the

contrary position that vowel reduction has two different mechanisms.

Acknowledging the bipartite nature of vowel reduction is key to explaining the empirical

facts of what I refer to as “reduction paradoxes”—cases in which vowel reduction patterns

indicate than one and the same vowel is both highly marked (i.e., tends to be subject to reduction

cross-linguistically) and highly unmarked (i.e., often serves as a reduction vowel, replacing other

vowel qualities that are subject to reduction). This sort of paradox can be resolved by

recognizing two types of constraints that focus on unstressed vowel qualities, but that have

separate teleologies. One type of constraint is based on the idea of prominence, and is

implemented using Prominence Reduction constraints (Prince and Smolensky 1993). With

respect to prominence-reducing vowel reduction, unstressed /a/ is disfavored, being a highly

sonorous vowel. The other is based on the idea of contrast, and is implemented using Licensing

constraints, specifically, Licensing constraints focussing on avoiding unstressed noncorner

vowels. In this sort of vowel reduction, unstressed /a/ is favored, since /a/ is one of the three

* I would like to thank the following individuals for valuable comments on this chapter, or on work leading to it:Tim Beasley, Bruce Hayes, Patricia Keating, Robert Kirchner, Joyce McDonough, and Donca Steriade.

2

corner vowels /i,u,a/. In what follows, I will lay out the constraints motivating these two types of

reduction, their phonetic motivations, and examples of how they work.

An additional point to be made is that the two constraint families alluded to above only

identify the vowels to be eliminated by vowel reduction and the contexts in which they are to be

eliminated. They do not, however, identify the method for eliminating them. Invoking these

constraints within Optimality Theory predicts that languages will vary in the exact

neutralizations used to meet the demands of these constraints, and furthermore that the set of

possible neutralizations corresponds to the logical possibilities predicted by a factorial typology

combining these reduction constraints with constraints on vowel faithfulness. Indeed, an

empirical survey of vowel reduction languages demonstrates a wide variety of vowel reduction

patterns, including multiple cases where the same unstressed sub-inventory is achieved via

different sets of neutralizations, and furthermore that the observed patterns are a good fit for a

factorial typology based on the analysis of vowel reduction provided here (Crosswhite 2001).

1.1 A Note on Vowel Features and Vowel Faithfulness

One vowel quality that shows up quite commonly in vowel reduction is [�]. I assume

that reduced [�] is essentially a targetless vowel that is not specified for any vowel features

(except perhaps, following Browman and Goldstein 1992, a specification for minimal opening).

For example, I assume that in English the weak, reduced vowel [�] and the full vowel [¥] differ

in featural representation: whereas [¥] is specified mid central, [�] is not. Assuming this

representation, I adopt a Dep/Max approach to vowel faithfulness (Zoll 1996). For example, the

reduction of /i/ to [�] would require the deletion of the feature specifications [+high, +front,

-low], in violation of the corresponding Max[F] constraints. Similarly, reduction to a non-schwa

vowel will be treated as the deletion of certain feature specifications, and the insertion of certain

3

others. For example, the reduction of /e/ to [i] would be represented as deletion of [-high]

(violating Max[-high]) and insertion of [+high] (violating Dep[+high]).

2. Contrast Enhancing Reduction

The first type of vowel reduction I will consider is contrast-enhancing reduction, in which

certain undesirable or perceptually-challenging vowel qualities are limited to stressed position.

In general, this form of reduction will amount to elimination of noncorner vowels, especially mid

vowels. One such reduction pattern is found in Belarusian (Krivitskii and Podluzhnyi, 1994), in

which the mid vowels /e,o/ both to reduce to [a]:

(1) Vowel Neutralizations in Belarusian (Krivitskii and Podluzhnyi, 1994)

Vowels Under Stress Same Vowels Unstressedªno¦i ‘legs’ naª¦a ‘leg’ªkol ‘pole’ (nom.) kaªla ‘pole’ (gen.)ªv-osni ‘spring’ (gen.) v-aªsna ‘spring’ (nom.)ªm-ot ‘honey’ (n.) m-aªdovi ‘honey’ (adj.)ª6ept ‘whisper’ 6apªtats- ‘to whisper’ªreki ‘rivers’ raªka ‘river’ªsp-ets- ‘to ripen’ paªsp-avats- ‘to mature’ªkl-ej ‘glue’ kl -aªjonka ‘oil-cloth’

This form of reduction produces an end result in which unstressed syllables are limited to

the vowel sub-inventory [i,u,a]. As noted by Lindblom (1986), this type of vowel inventory

shows maximal dispersion and, therefore, minimal acoustic ambiguity. This and similar types of

vowel reduction will be motivated using Licensing constraints, which are discussed below.

2.1 Licensing Constraints and Contrast Enhancement

Not all speech sounds are perceived equally well, furthermore, not all speech sounds are

equally good in all segmental or prosodic environments. From the speaker’s point of view, it

may be undesirable for a speech sound to be misperceived—not merely out of charitable concern

for the listener, but also out of selfish reasons: If you produce a speech sound that is

4

misperceived, you have expended articulatory effort in an ineffective manner. This approach is

the basis for Steriade’s (1994a,b) licensing-by-cue approach to phonological neutralizations: If a

given contrast is in danger of being missed by the listener, why should the speaker go to the

trouble of producing it? In other words, positional neutralizations based on the desire to avoid

ineffectual expenditure of articulatory effort can be thought of as the grammatical encoding of

the speaker’s preference to "not deploy a feature in positions where its defining [acoustic] cues

are necessarily absent or diminished" (Steriade 1994a). There are two logical courses of action

for a speaker who wants to avoid ineffectual articulation: (1) Don’t say the sound at all, or (2)

Say a different, but similar, sound. In this study, I do not consider strategy 1 (deletion), instead

focussing on strategy 2 (neutralization).

To account for contrast-enhancing vowel reduction, I will use Licensing constraints—a

non-faithfulness-based version of Steriade's (1994a) Implement constraint family. (See also

Steriade's (1994b) Positional Neutralization constraints, as well as the Stress-Prominence

constraints used by Majors (1998) in the analysis of stress-dependent vowel harmony.) The

form of a Licensing constraint is as follows:

(2) LIC-Q /β: The vowel quality Q is only licensed in context β.Where Q = any vowel quality or a natural vowel class

β = any context that enhances the accurate perception of Q

Note that a Licensing constraint cannot combine just any vowel quality Q with just any context

β: Licensing is constrained by the requirement that the context β must enhance the accurate

perception of Q. It should also be underscored at this time that Licensing constraints are not

members of the Faithfulness constraint family. A constraint such as LIC-Q/β will assign a

violation mark for every instance of [Q] that occurs without β, irrespective of whether [Q] is

underlying or derived.

5

In this respect, Licensing constraints are similar to the grounding conditions discussed by

Archangeli and Pulleyblank (1994). For example, drawing on the fact that tongue root

advancement and tongue body raising are articulatorily compatible gestures while the

combination of tongue root advancement and tongue body lowering are articulatorily

antagonistic, they posit constraints of the type ATR/Low, which prohibits the [-ATR] feature

specification from co-occurring with a [-low] specification (that is, vowels that are [-ATR] must

be low). The Licensing constraints used here take a similar approach, applying not to

combinations of features (as in ATR/Low), but to combinations of features with positions (such

as stressed position). It should also be pointed out that Archangeli and Pulleyblank’s motivation

for the ATR/Low constraint (and, indeed, for most of their grounding conditions) is based on

articulatory considerations: retraction of the tongue root allows easier depression of the tongue

body—thus [-ATR] and [+low] are articulatorily compatible. Archangeli and Pulleyblank also

allow grounding conditions to refer to acoustic compatibility of the type discussed with respect

to Licensing above, but their emphasis is usually on articulatory compatibility.1 In the

enhancement-based Licensing constraints used here, articulatory considerations do not play a

role, although acoustic compatibility is required (the licensing context β must enhance correct

perception of Q).

The particular licensing constraints considered here are those focussed on particular

vowel qualities in stressed position. The licensing constraint that motivates most cases of

contrast-enhancing vowel reduction is:

(3) LIC-Noncorner/Stress: Noncorner vowels are licensed only in stressed positions.

In order to maintain the phonetic motivation for contrast-enhancing vowel reduction, it is

necessary to demonstrate two facts: (1) that noncorner vowels (the quality Q) are subject to

6

misperception, and (2) that stressed position (the context β) enhances their correct perception.

These two issues are addressed separately below.

2.2 Corner vs. Noncorner Vowels

The corner vowels /i,u,a/ show several special properties: they are the three most

common vowels, occurring in almost all languages (Maddieson 1984); and, as a set, they

constitute the smallest complete vowel inventory found with any regularity in the world's

languages (Maddieson 1984, Lindblom 1986). Although much remains to be learned about these

vowels, three characteristics work to single out /i,u,a/ from all remaining vowel qualities:

dispersion (Lindblom 1986), quantal characteristics (Stevens 1986) and focalization (cf. Stevens

1986, Schwartz et al. 1997).

Dispersion refers to the efficient use of the acoustic space available: Speech sounds

should be well dispersed throughout this space, so as to increase the distinctiveness of each of

the sounds from the others. It is for this reason that certain vowel inventories are very common

across the world's languages: they locate vowels at various points in acoustic space that

maximize the acoustic distance between the members of the inventory. This idea, often referred

to as Dispersion Theory, has been articulated in a number of articles by Bjorn Lindblom and

collaborators (Liljencrants and Lindblom 1972, Lindblom 1986, Lindblom and Maddieson 1988,

etc.) Under this theory, the corner vowels /i,u,a/ are special in that they are maximally

acoustically distinct: In theory, a vowel system consisting of these three vowels would be easiest

in terms of perception because the possibility for confusing an intended vowel quality for an

incorrect but adjacent vowel quality is minimized. However, this cannot be the only motivation

for treating the corner vowels /i,u,a/ as special. As illustrated in Lindblom (1986), the distance

metric alone does not uniquely identify /i,u,a/ as having this quality. The vowel inventory /i,u,o/

7

also displays maximal dispersion (and, as pointed out by Lindblom, is an attested 3-vowel

inventory).

The vowels /i,u,a/ are also special is in terms of their production: they all show quantal

effects (Stevens 1986). A quantal effect occurs when a given change in articulation does not

produce a correspondingly large acoustic change. That is, speech sounds that show quantal

effects are ones in which the appropriate acoustic quality is more or less consistent with a wide

range of articulations. Non-quantal sounds, in contrast, show large changes in acoustic quality

for similarly-sized articulatory changes. Quantal Theory (Stevens 1986) hypothesizes that

languages prefer to use speech sounds that show these quantal effects, presumably because they

are consistent with a wider range of articulations, and thus easier to produce under a wide range

of contexts. In particular, the vowels /i,u,a/ all show this sort of effect. For example, as

illustrated by Perkell and Nelson (1982, 1985) and Perkell and Cohen (1986), the articulation of

/i,u,a/ do show significant variation in constriction location, but not in terms of constriction

degree, supporting the idea that these vowel qualities are fairly stable in the face of certain types

of articulatory changes as identified by Quantal Theory. However, it is again the case that

quantal effects alone are insufficient to explain why /i,u,a/ constitute a special class: it is not the

case that /i,u,a/ are the only vowels that show such effects. Indeed, it has been suggested that

central vowels like /¥/ an /�/ also show quantal effects (Ladefoged et al. 1977, Pisoni 1980).

However, as pointed out by Diehl (1986), the vowels /i,u,a/ are distinguished by having not only

articulatory stability, but also in the fact that they do not share this characteristic with adjacent

qualities. That is, the vowels /i,u,a/ not only occupy areas of acoustic stability in the face of

articulatory variation, but they are surrounded by areas of instability, where acoustic quality is

comparatively sensitive to articulatory variation.

8

Furthermore, Stevens (1986) also points out that the quantal vowels /i,u,a/ and /y/ are

perceptually special by virtue of having spectral prominences caused by convergences: either

proximity of two formants (/i/: F3, F4; /a/: F2, F1; /y/: F2, F3), or proximity of the first formant

and the fundamental frequency (/u/). Stevens suggests that these proximities are (or tend to be)

less than some critical distance necessary for distinguishing both prominences; i.e., that the two

formants (or F1 and FØ) are so close in these vowel qualities as to converge auditorily into a

single spectral prominence.

Schwartz et al. refer to the presence of such convergences as giving a vowel a "focal"

property. Based on the finding that these focalizations lead to more stable patterns in

discrimination tasks, Schwartz et al. hypothesize that they also lead to increased perceptual

salience. For an intuitive explanation of this effect, they quote Lieberman (1971:57-58) as

saying that these vowels "provide acoustic salience: that is, their formant frequency patterns

yield prominent spectral peaks (formed by the convergence of two formant frequencies […]) that

make is easier to perceive the sounds, just as, in the domain of color vision, saturated colors are

easier to differentiate than muted ones." Schwartz et al. then propose a hybrid Dispersion-

Focalization Theory, which seeks to maximize not just distance between vowels (inter-vowel

salience, dispersion), but also the intra-vowel salience or "local focalization." Under this

approach, the possible three-vowel inventories /i,u,a/ and /i,u,o/ (both of which maximize

dispersion) are distinguished, with /i,u,a/ being more optimal since it maximizes both dispersion

and focalization.

Given the above, it does seem clear that the vowels /i,u,a/ do have a special status. This

status does not seem to be linked to any single characteristic, but perhaps to the amalgamation of

several, making this set of vowels particularly auspicious: Their qualities are stable against

articulatory variation, they are maximally acoustically distinct from one another, and they are

9

perceptually salient due to convergences of spectral prominences. Not only are they individually

"good" vowel qualities, but as a set, they also constitute a particularly desirable vowel inventory.

It therefore seems reasonable to suppose that precisely these vowels would be the most desirable

to use in cases where correct vowel perception is at risk. Put another way, vowels that are not

members of this set might reasonably be limited, in some languages, to those positions where

their correct perception is most favorable. As such, this set of vowels seems like a plausible

candidate for the quality Q mentioned in a Licensing constraint—that is, the set that requires a

particular environment to be phonologically licensed. With this in mind, I now turn to consider

the other side of the question, namely the position where these vowels are licensed.

2.3 Motivating Licensing Under Stress

As mentioned earlier, this section specifically focuses on cases where noncorner vowels

are neutralized in unstressed syllables. As just outlined, corner vowels have several special

characteristics that make them particularly desirable phonetically. However, is it the case that

stressed position is a context that would increase the correct perception of a noncorner vowel? In

general, increasing the exposure to any stimulus increases the likelihood that the stimulus will be

correctly identified. Assuming that stress engenders increased duration, the condition placed on

Licensing constraints (β must enhance correct perception of Q) is met, and stressed position is an

appropriate context for a licensing effect.

In connection with this, it is interesting to note that the constraint proposed, Lic-

Noncorner/Stress, appears to be active only in those languages where stress is correlated with

increased duration: I have found no examples of stress-induced licensing of vowel features that

occur in either languages where stress is strictly intensity-based (i.e., stressed vowels are not

longer than unstressed ones, as in contemporary Czech (Palková 1994)), or in languages that use

pitch-accent. There are, however, interesting examples where some variety of a language that

10

predominantly shows pitch-accent or intensity-based stress does possess vowel reduction. In all

such cases I have found, the accentual system of the innovating dialect has replaced the prosodic

system of the predominating dialects with a duration-based stress system. Examples include

Standard Slovene (Bezlaj 1939, ToporiÎi� 1976, Len�ek 1982), Horjulj Slovene (Len�ek

1982:145), Botzetierra Basque (Hualde 1991) and dialectal Polish (Uba¸czyk, 1953:11). There

are also cases that go the other way around, in which a language shows predominantly duration-

based stress, but also has dialects that lack vowel reduction. In these cases, the dialects in

question lack a strong duration-based stress (cf. Russian with okan’e, cf. Kasatkin 1989).

2.4 Effect of Lic-Noncorner/Stress

The Lic-Noncorner/Stress constraint induces elimination of unstressed mid vowels. In

the Belarusian example provided earlier, both the unstressed mid vowels /e,o/ were eliminated

via lowering to [a]. However, this is not the only possible result for Lic-Noncorner/Stress. For

example, the exact opposite phenomenon—reduction via raising—is also attested. An example

of this reduction pattern is found in the Native American language Luiseño (Munro and Benson

1973):

(4) Vowel Neutralizations in Luiseño (Munro and Benson 1973)

Vowels Under Stress Same Vowels Unstressedunstressed ªt6oka ‘to limp’ t6uªkat6ka6 ‘limping’/e,o/ raise ªhedin ‘will open’ hiªdiki- ‘to uncover’

t6aªpomkat ‘liar’ ªt6a6pumkatum ‘liars’unstressed ªmaha ‘to stop’ maªhamha6 ‘slow’/i,u,a/ do ªku·mit ‘smoke’ kuªmikmi6 ‘smoky colored’not reduce ª�u·kat ‘deer’ ªpa·�ukat ‘elk’

taªkitki6 ‘straight’ ªta·ki6 ‘stone forsmoothing pottery’

In the Luiseño case, unstressed mid vowels are eliminated by raising. Note that the

remaining vowels do not undergo reduction, once again creating the maximally dispersed vowel

11

sub-inventory [i,u,a] in unstressed positions. The difference between Belarusian and Luiseño

can easily be accounted for by using different rankings for vowel faithfulness constraints:

Belarusian eliminates unstressed mid vowels in a manner that preserves their underlying non-

high nature, while Luiseño eliminates the same vowels in a manner that preserves their

underlying color. The following partial grammars illustrate this state of affairs:

(5) Partial Grammars for Belarusian and Luiseño Vowel Reductions

Belarusian:Lic-Noncorner/StressMax[-high]

» Max[round],Max[+front]

Luiseño:Lic-Noncorner/Stress,Max[round], Max[+front]

» Max[-high]

In both the Belarusian and Luiseño cases, the Lic-Noncorner/Stress constraint is undominated (at

least in these partial grammars), thus motivating the elimination of unstressed /e,o/. In addition,

in each case at least one, if not two, vocalic faithfulness constraints are also undominated. In the

case of Belarusian, the undominated faithfulness constraint is Max[-high], which derives

reduction-via-lowering, as illustrated below for two hypothetical forms:

(6) Reduction via Lowering (e.g., Belarusian)

/toªta/‘hypothetical word’

Lic-Noncorner Max[-high]

Max[round]

Max[+front]

) [taªta] *

[tuªta] *!

[toªta] *!

/teta/‘hypothetical word’

Lic-Noncorner Max[-high]

Max[round]

Max[+front]

) [taªta] *

[ti ªta] *!

[teªta] *!

12

As illustrated, the two topmost constraints do not need to be ranked with respect to one

another—the fact that the color-based faithfulness constraints are at the bottom of the constraint

hierarchy is adequate to derive reduction via lowering. By reversing the position of the color-

based faithfulness constraints and Max[-high], reduction via raising is derived:

(7) Reduction via Raising (e.g., Luiseño)

/toªta/‘hypothetical word’

Lic-Noncorner Max[round]

Max[+front]

Max[-high]

) [tuªta] *

[taªta] *!

[toªta] *!

/teta/‘hypothetical word’

Lic-Noncorner Max[round]

Max[+front]

Max[-high]

) [tiªta] *

[taªta] *!

[teªta] *!

Again, the only ranking that need be specified to derive reduction via raising is that Max[-high]

must be at the bottom of the constraint hierarchy. Other logically possible permutations of these

constraints predict yet other neutralization patterns for eliminating unstressed mid vowels, all of

which seem to be attested. For example, by ranking Max[round] at the bottom of the constraint

hierarchy, an asymmetrical reduction pattern (/e/ > [i], /o/ > [a]) is predicted, as shown below.

This type of pattern is attested in Contemporary Standard Russian.

(8) An Asymmetrical Reduction Pattern (e.g., Russian)

/toªta/‘hypothetical word’

Lic-Noncorner Max[+front]

Max[-high]

Max[round]

) [taªta] *!

[tuªta] *

[toªta] *!

13

/teta/‘hypothetical word’

Lic-Noncorner Max[+front]

Max[-high]

Max[round]

) [tiªta] *

[taªta] *!

[teªta] *!

Reversing the position of Max[+front] and Max[round] predicts the opposite asymmetrical

pattern: /e/ > [a], /o/ > [u]. This pattern is attested in Algueres Catalan (Recasens 1991).

Similarly, by adding additional vocalic faithfulness constraints, even more variations on the

elimination of unstressed mid vowels are accounted for. Some of these additional patterns

include [ATR]-preserving patterns (cf. Bergün Romansch: /(,o/ > [a], /e/ > [i], /o/ > [u], Lutta

1923, Kamprath 1991), and [low]-preserving patterns (cf. Saipanese Chamorro: /4�$/ > [a]).

For a more detailed description of predicted reduction patterns and their attestation, see

Crosswhite (2001).

2.5 Licensing-based Reduction Creating Novel Vowel Qualities

Another type of licensing-based reduction creates as its output vowels that do not belong

to the basic inventory of vowel phonemes. We consider first a case from Slovene. In Slovene,

there are 8 phonemic vowel qualities: /i,u,e,o,(,o,a,�/. Note the phonemic contrast between lax

and tense vowels, as illustrated by pairs such as [ªtseÛsta] ‘road’ vs. [ªs(Ûstra] ‘sister’. However,

this opposition is only maintained in long, accented syllables. Since stress is mobile in Slovene,

this produces tense~lax alternations, as illustrated below (data from Bidwell 1969):

14

(9) Vowel Neutralizations in Standard Slovene (Bidwell 1969)

Vowels Under Stress Same Vowels Unstressedunstressed o ªgooÛÛra ‘mountain’ nom. sg. gooªreÛ ‘mountain’ gen. sg.(no change) ªpooÛÛtok ‘stream’ nom. sg. pooªtoÛka ‘stream’ gen. sg.unstressed ( ªpl((ÛÛma ‘tribe’ nom. sg. pl((ªmeÛna ‘tribes’ nom. pl.(no change) ªs((ÛÛstra ‘sister’ nom. sg. s((ªstreÛ ‘sister’ gen. sg.unstressed o > o ªmo= ‘man’ nom. sg. mooª=jeÛ ‘men’ nom. pl.

ªkoÛÛst ‘bone’ nom. sg. kooªstiÛ ‘bone’ gen. sg.unstressed e > ( ªreÛÛt6 ‘word’ nom. sg. r((ªt6iÛ ‘word’ gen. sg.

ªtseÛÛsta ‘road’ nom. sg. ts((ªsteÛ ‘road’ gen. sg.

The tense~lax distinction is also neutralized in short, accented syllables, as illustrated by forms

such as [ªkm((ÛÛta] ~ [ªkm((t] ‘peasant’ (gen./nom.) As illustrated, the traditional description is that

when the tense~lax distinction is neutralized, it is in favor of the lax mid vowels [(,o]. The only

other known case of reduction via laxing is from certain North-Eastern dialects of Brazilian

Portuguese (Brakel 1985, Perrone and Ledford-Miller 1985). The existence of reduction via

laxing is something of an anomaly, since the opposite form of reduction— reduction via tensing

(see section 0)—seems much more common, and moreover fits in with the theoretical

approaches to vowel reduction presented here. However, instrumental analysis of the Slovene

tense~lax neutralization by Lehiste (1961) indicates that there is more to the story than would be

guessed based on examining traditionally-transcribed forms, such as those presented above. She

finds that the neutralized vowels usually transcribed as [(] and [o] are actually intermediate

between non-neutralized /e/ and /(/ or /o/ and /o/. Lehiste therefore proposes that the Slovene

tense~lax neutralization is actually a case of archiphonemic neutralization: the neutralized

vowels are simply unspecified for laxness or tenseness and might, therefore, be more accurately

transcribed as the archiphonemes [E,O]. The following graph, based on the vowel formant

measurements reported by Lehiste (1961), illustrates this situation:

15

(10) Stressed and Unstressed Mid Vowels in Slovene (data from Lehiste 1961)

This is in stark contrast with reduction via tensing. For example, an underlying tense~lax

distinction is also neutralized in many dialects of Italian, but this time in favor of the tense

vowels. Data illustrating this phenomenon are provided below:

(11) Vowel Neutralizations in Standard Italian

Vowels Under Stress Same Vowels Unstressedunstressed o > o akªkoo��ere ‘to receive’ akko�ª�(nte ‘welcoming’

ªfoo��o ‘page’ fo�ª�etto ‘slip of paper’unstressed ( > e ªb((lla ‘beautiful’ belªlettsa ‘beauty’

ªp((dd=o ‘worse’ pedªd=ore ‘to worsen’unstressed o ªsole ‘sunlight’ soledªd=ato ‘sunny’(no change) ªbokka ‘mouth’ bokªka��o ‘mouthpiece’unstressed e ªpelo ‘hair’ peªloso ‘hairy’(no change) ªrete ‘net’ reªticolo ‘network’

Instrumental analysis by Baroni (1996) demonstrates that the neutralized vowels in Italian are

not intermediate between the qualities observed for the stressed lax and tense mid vowels.

Rather, the neutralized vowels (transcribed as unstressed [e,o] above) are in fact qualitatively

indistinguishable from stressed [e,o]. Thus, we have a contrast: a tense~lax distinction is

eliminated in unstressed syllables in Italian by replacement (lax vowels are replaced by tense

16

ones), while in Slovene a lax~tense distinction is eliminated in unstressed (and short stressed)

syllables by feature elimination (both lax and tense vowels are replaced by an archiphoneme not

specified for [ATR]).

A case that is formally similar to the Slovene tense~lax neutralization is found in Eastern

Ojibwa (Bloomfield 1957, Miller 1972). In Eastern Ojibwa, there are basically three

monomoraic vowel qualities: /,� 8, a/. In unstressed positions (described by Bloomfield 1957

simply as odd syllables counting from the left edge of the word), /,/ and /a/ reduce to [�], while

/8/ reduces to a rounded variant of schwa, [��]. Superficially, this pattern resembles reduction of

all vowels to [�] (see section 3.7); however it differs in that one distinctive feature, namely

[round], is preserved under reduction. Recall here that [�] is analyzed as a featureless vowel.

Therefore [��] would be represented as a surface vowel with a specification for [round], but no

specification for height or advancement.

The Licensing effects discussed in the preceding sections can be described as bans on

certain vowel qualities in unstressed positions, leading to the elimination of certain vowel

specifications. However, both the Slovene and Eastern Ojibwa cases could be modeled by

banning certain vowel distinctions in a given position, leading to the wholesale absence of some

feature in that position. In the case of Slovene, the feature [ATR] cannot occur in monomoraic

positions. In the case of Eastern Ojibwa, the features [high] and [front] do not occur in

monomoraic unstressed positions. This might suggest Licensing constraints such as the

following:

(12) Lic-[ATR]/µµ: Feature specifications for [ATR] may only occur in association withbimoraicity.

(13) Lic-[high, front]/Stress: Feature specifications for [high] and/or [front] may only occur inassociation with stress.

17

This sort of Licensing constraint can be seen as establishing a second vowel inventory for

unstressed and/or short positions by redefining the set of features allowable in these contexts.

That is, this sort of Licensing constraint actually introduces new vowel qualities that occur in

unstressed or short positions, but are absent from stressed or long positions. This apparent

increase in vocalic complexity dependent on stress/length is seemingly mitigated by the fact that

the unstressed/short vowel inventory uses fewer features, and therefore makes fewer vowel

distinctions, despite the fact that it introduces additional vowel qualities.

3. Prominence Reduction

In the preceding section, I discussed a number of vowel reduction systems based on the

idea of avoiding noncorner vowels in unstressed position. In this type of system, the unstressed

vowel [a] was considered to be highly desirable. Interestingly, there also vowel reduction

systems that seem to suggest exactly the opposite conclusion: in these systems, [a] is illegal in

stressless position, and reduces to [�]. A resolution of this paradox, as we will see, can be found

in enriching the constraint system, attributing constraints with conflicting effects to conflicting

phonetic teleologies.

One example of a phonological system that reduces stressless [a] to schwa is found in

Bulgarian. As in various systems discussed so far, the noncorner vowels /e,o/ are neutralized in

unstressed syllables, surfacing as [i,u], respectively, as shown in (14a-c). However, the vowel

/a/ is reduced to [�] (14d-e). Clearly, this reduction pattern is at odds with the phenomenon of

contrast-enhancement.

18

(14) Vowel Reduction in Bulgarian

Vowels Under Stress Same Vowels Unstresseda. ªroguf ‘of horn’ ruªgat ‘horned’b. ªonzi ‘that’ (masc.) uªnazi ‘that’ (fem.)c. ªselu ‘village’ siªla ‘villages’d. ªrabut� ‘work’ r ��ªbotnik ‘worker’e. ªgrat ‘city’ gr ��ªdets ‘town’

A further fact, which fits into this general pattern, is that Bulgarian also has a phonemic /�/,

which does not undergo neutralization: unstressed /�/ emerges without change. This only adds

to the puzzle: Since the vowel quality [�] is at least as nonperipheral as the qualities [e,o] (if not

more so), why does this vowel escape reduction? To solve this type of vowel reduction paradox,

I propose that Bulgarian vowel reduction, and similar reduction types, constitute a formally

distinct type of vowel reduction, which I refer to as prominence reduction. Prominence-

reducing vowel reduction is based on the desire to avoid particularly long or otherwise salient

vowel qualities in unstressed positions. A wide variety of vowel reduction patterns seem to fall

into this category. This type of vowel reduction will be modeled here using Prominence

Alignment constraints, as discussed by Prince and Smolensky (1993).

3.1 Prominence Alignment

The formal mechanism of Prominence Alignment was first employed by Prince and

Smolensky (1993) in an analysis of predictable syllabicity in Imdlawn Tashlhiyt Berber, in

which any segment can be syllabic, and the choice of which individual segments will surface as

syllable nuclei in any given word is predicted by relative sonority. Specifically, more sonorous

segments are chosen to occupy the syllable nuclei positions, while less sonorous segments are

chosen to occupy syllable onset positions. Prince and Smolensky analyze this pattern as a case

of "stacking," or aligning, prominent elements. Assuming that syllable nucleus position is a

19

prominent prosodic position, and furthermore that increasing segmental sonority is correlated

with increasing prominence, the Imdlawn Tashlhiyt Berber pattern can be accounted for simply

by saying that prominence at both the segmental and syllabic levels should co-occur. That is,

combinations of prominent syllabic position plus non-prominent segmental material (or vice-

versa) should be avoided. This is formalized by Prince and Smolensky starting with

identification of two "phonetic scales." The two scales they used are illustrated below. (The

symbol " prom>" means "is more prominent than.")

(15) Scale 1: Syllabic prominencepeak prom> margin

Scale 2: Segmental prominence (sonority)a prom> e,o prom> i,u prom> l,r prom> n,m prom> etc.

These two "phonetic scales" are then crossed, as follows: Choose one element of Scale 1 and

cross it, in order, with each member of Scale 2. The resulting combinations form the basis of

Optimality-Theoretic constraints militating against that particular combination of elements.

These constraints are inherently ranked with respect to one another, mirroring the order of

elements in Scale 2.

Since Scale 1 has two members, there are two different ways the crossing operation can

proceed, producing constraints that either focus on the prominent member of Scale 1, or the non-

prominent member of Scale 1. The two constraint families that can be made by crossing the two

scales shown in (15) are illustrated below:

(16) *Margin/a » *Margin/e,o » *Margin/i,u » *Margin/l,r »*Margin/n,m » …(*Margin/X = X is not a syllable margin.)

(17) … » *Peak/n,m » *Peak/l,r » *Peak/i,u » *Peak/e,o » *Peak/a(*Peak/X = X is not a syllable peak.)

Note that the constraints in (16) focus on the non-prominent member of Scale 1, namely, syllable

margin position. Following Jian-King (1996), I will refer to this type of constraint family as a

20

Prominence Reduction family. Note that a Prominence Reduction family has members ranked in

order of decreasing sonority, encoding the idea that reductions of prominence are preferred in

non-prominent positions. For example, the fact that *Margin/a is the highest-ranked member of

this family expresses the idea that the vowel quality [a] is in fact the worst possible syllable

margin imaginable—the fact that it is the highest-ranked constraint means that it is the most

difficult to violate. Similarly, the constraints in (17) focus on the prominent member of Scale 1,

namely, syllable nucleus position. I will refer to this type of family as a Prominence Alignment

family. Note that its members are ranked in order of increasing sonority, expressing the idea that

high prominence is preferred in prominent positions. For example, the fact that *Peak/a is the

lowest-ranked member of this family expresses the idea that [a] is in fact a very good syllable

nucleus—the fact that it is the lowest-ranked member means that it is the easiest one to violate.

The idea of crossing two phonetic scales is not limited to sonority and syllabicity. In

theory, any two prominence scales can be crossed in a similar way. For example, Kenstowicz

(1994) employs this technique to cross sonority with stress (stressed prom> unstressed), explaining

why certain high-sonority vowels in some language attract stress, while certain low-sonority

vowels in other languages repel stress. Similarly, Gordon (this volume) uses constraints crossing

stress with rime prominence to explain patterns of syllable weight, and Crosswhite (1999)

crosses sonority with moraicity to explain why certain high-sonority vowels undergo lengthening

in stressed position more easily than low-sonority vowels.

3.2 Motivating Prominence Alignment

Prince and Smolensky (1993) make it clear that the Prominence Alignment mechanism is

intended to be phonetically-motivated: the inherent ranking of constraints in these families is

intended to mirror some specific physical continuum that could, presumably, be objectively

established through measurement. However, the exact physical continuum appropriate for this is

21

far from clear. For example, it has been noted several times that the phonological sonority

hierarchy cannot straightforwardly be reduced to any single phonetic characteristic. For

example, Fry (1979) attempts to correlate phonological sonority in English with intensity (louder

= more sonorous), while Lindblom (1983) suggests a basis for sonority in jaw opening (more jaw

depression = greater sonority). However, both of these proposals fail to generate the precise

ordering employed by phonologists in analyses of, say, sonority sequencing in syllabification.

The major problems with these approaches are laid out by Keating (1983) and Malsh and Fulcher

(1989). For example, although segments can be classed by their preferred degree of jaw opening

(as in Lindblom 1983), it turns out that most consonants can easily accommodate a wide range of

jaw positions, while relatively few segments (such as /s/) have a more demanding, less variable

jaw position (Keating 1983). Furthermore, although intensity correlates with sonority fairly well

for sonorant segments like vowels, liquids, and nasals, defining sonority in terms of intensity

makes incorrect predictions concerning obstruents: Certain obstruents, such as fricatives, have a

fairly high intensity due to the presence of noise produced at a constriction site somewhere in the

vocal tract, yet they are considered to be low-sonority segments.

In response to these difficulties, some researchers have proposed that there is not a single

physical correlate for sonority, but many. For example, Malsh and Fulcher (1989) propose that

sonority is correlated to intensity plus jaw opening—under their framework, segments that are

both loud and open (such as vowels and sonorous consonants) will be highly sonorous, while

segments that are both quiet and close (such as stops) are highly non-sonorous. Segments that

are ambiguous, such as /s/, which has a close jaw position but high intensity, will be ordered on a

language-specific basis. In a similar tack, Nathan (1989) proposes a number of physical

correlates (voicing, openness, "prolongability") that tend to make a segment more prototypically

sonorous. However, I propose that is too early to give up hope for a straightforward physical

22

correlate for sonority. In connection with this, it is worth noting a similar situation regarding

another prominence-based phonological phenomenon, syllable weight. As demonstrated by

Gordon (this volume), syllable weight hierarchies are in some cases straightforwardly correlated

with simple duration measurements, but this correlation is imperfect—a slightly different

measure, total perceptual energy, provides a better basis for predicting syllable weight

distinctions. With this in mind, it may be the case that some other, more specific, measure may

provide a phonetic basis for the phonological sonority hierarchy. For example, in Keating's

(1983) discussion of sonority, she points out that jaw opening and the frequency value for the

first formant (F1) are both correlated with amplitude at low frequencies: lower jaw position

correlates with a higher F1 and higher low-frequency amplitude. In contrast, noise associated

with obstruents tends to be broad-spectrum or high-frequency. Similarly, Stevens (1986:35-37)

notes that sonorant and non-sonorant consonants differ in low-frequency amplitude: producing a

constriction somewhere in the vocal tract in turn decreases transglottal pressure and causes a

decrease in the amplitude of glottal pulses. According to Stevens, this causes a drop-off in

amplitude at low frequencies, specifically in the vicinity of the first harmonic. These

observations, coupled with the fact that the human auditory system responds differently for low-

and high-frequency sounds (those above or below approximately 3,000 Hertz; Johnson 1980), it

seems at least reasonably plausible that sonority should not be equated with general amplitude,

but with low-frequency amplitude.

Although this hypothesis has yet to be tested experimentally, it is clear that this

hypothesis, or some form of it, satisfactorily accounts for vocalic prominence. Indeed, it may be

the case that vocalic prominence and consonantal prominence are not truly the same creature: in

determining the phonetic basis for phonological sonority hierarchies, it is generally the ordering

of obstruents that is problematic. For example, Wright (this volume) makes a compelling case

23

for the role of perceptual robustness in accounting for consonantal sonority sequencing

restrictions. In comparison with obstruents, the relative sonority of vowels and sonorant

consonants is relatively straightforward. For example, it has been found for a number of

languages that low vowels tend to be longer in duration than mid vowels, which in turn tend to

be longer than high vowels (cf. Lehiste 1970 for overview). Lehiste (1970) proposes that this

low > mid > high duration pattern is in fact universal, and rooted in physiological factors. That

is, vowels generally require a fairly open vocal tract, and one way of achieving this is through

jaw depression. The lower the vowel, the more jaw depression is typically observed, and hence,

a longer articulation time. Furthermore, the more open the vocal tract is, the more sound can

escape, thus increasing intensity. Although this phonetic explanation does not extend to

consonants, it is clear that vocalic prominence is straightforward: the longer and louder a vowel

is, the more prominent it is. By virtue of the way in which vowels are articulated, both of the

characteristics are generally correlated with jaw depression. This definition of vocalic

prominence has the added benefit that it can easily be extended to new vowels. For example,

Kenstowicz (1994) proposes extending the sonority hierarchy to include [�] as the least sonorous

of vowels to explain why, in certain languages, this vowel has the special property of repelling

stress. Taking the common assumption that [�] is a mid vowel, this extension is unmotivated: If

sonority were correlated simply to tongue height, we would expect [�] to have a medium

sonority, similar to that of [e,o]. However, under the current hypothesis, this extension is

expected: For example, Gruzov (1960) explores vowel duration in Mari, a language where [�]

does have the stress-repelling property investigated by Kenstowicz (1994). His data show that

this vowel is indeed significantly shorter than the other vowels not showing this behavior, a

characteristic which extends even to cases of stressed [�], where lengthening under accent is

24

minimal. Similarly, jaw depression measurements for [�] in Bulgarian indicate that this vowel is

very close, showing a jaw depression about the same as for Bulgarian [i,u].2

With this interpretation of vocalic prominence in mind, prominence-stacking effects such

as low-sonority vowels repelling stress (Kenstowicz 1994), or low-sonority vowels lowering

under stress (Crosswhite 1998), can be seen as a grounded effect, in the sense of Archangeli and

Pulleyblank (1994). That is, the only characteristic that seems to be held in common between

vocalic prominence (sonority) and positional prominence (such as stress or syllabicity) is the

presence of either increased intensity, increased duration, or both. By making prominent

positions co-occur with sonorous segmental material, entities with the same or similar phonetic

characteristics, in this case, duration and intensity, are made to co-occur within a single

segmental locus. This goal can be realized in a number of ways. The specific case of

prominence-stacking investigated by Prince and Smolensky was syllabification—in their

examples, syllable nuclei were chosen based on their relative sonority, with the most sonorous

segments preferentially parsed as syllabic. For example, a low-sonority vowel like [i] would be

parsed as an onset or coda if doing otherwise would require a higher sonority vowel like e.g. [o]

or [a] to be non-syllabic. In other words, the underlying sonority of the segments involved

remains static, and prosodic prominence (i.e., syllabicity) is manipulated to provide a good

match. However, the opposite is also possible: the sonority of the underlying segments can be

increased in order to make them into better syllable nuclei. For example, in the (now-extinct)

Native American language Gabrielino (Munro, p.c.), short high vowels never surface as syllable

nuclei: When such a form would otherwise be expected (i.e., when underlying /iÛ/ shortens in

unaccented position), the vowel lowers to mid. A similar phenomenon occurred historically in

most Slavic dialects (the so-called ‘fall of the jers’, cf. e.g. Vaillant 1958), in which the short

25

high vowels /i,u/ were either deleted, or lowered to a more sonorous vowel quality (usually to

[e,o], although the exact vowel qualities vary from dialect to dialect). In both of these cases, the

underlying sonority of vowels is manipulated in order to provide a good match for prosodic

prominence. It is precisely this sort of case—the reduction of underlying sonority in unstressed

positions—that is discussed in more detail below.

3.3 Applying Prominence Alignment/Reduction to Vowel Reduction

Prominence-reducing vowel reduction results from the application of Prominence

Reduction constraints to vowels in unstressed positions. For example, the following constraint

hierarchy is found in several languages with prominence-reducing vowel reduction.

(18) *Unstressed/a » *Unstressed/(,o » *Unstressed/e,o » *Unstressed/i,u » *Unstressed/�

This constraint hierarchy is calculated by crossing the following two phonetic scales:

(19) Scale 1: Accentual prominencestressed prom> unstressed

Scale 2: Vocalic prominencea prom> (,o prom> e,o prom> i,u prom> �

As before, the Prominence Reduction constraint family that would be based on these scales is

produced beginning with the non-prominent member of Scale 1 (unstressed position). This is

systematically crosses it with the members of Scale 2, starting with the most sonorous member

[a]: The highest ranking Prominence Reduction constraint based on these two scales is therefore

*Unstressed/a (‘low vowels are not found in unstressed position’).

To demonstrate how Prominence Reduction applies to unstressed vowels in more detail,

two examples of prominence-reduction vowel reduction are discussed below: Bulgarian and Sri

Lankan Portuguese Creole.

26

3.4 Prominence Reduction in Bulgarian:

In Standard Bulgarian (as well as many Bulgarian dialects), there is a 6-vowel inventory:

/i,u,e,o,�,a/. Note that /�/ is a phonemic vowel in Bulgarian, which can occur both stressed and

unstressed. In most of the eastern dialects of Bulgarian, this 6-vowel inventory reduces to the 3-

vowel sub-inventory [i,u,�] when in unstressed position. This is accomplished via step-wise

raising of unstressed non-high vowels: unstressed /a/ raises to [�], while unstressed /e/ and /o/

raise to [i] and [u], respectively. These neutralizations are illustrated below.3

(20) Bulgarian Vowel Neutralizations

i uÇ Ç

e � oÇ

a

Forms illustrating the Bulgarian vowel reduction pattern are provided in (21) (repeated here for

the reader’s ease from (14)).

(21) Bulgarian Vowel Reduction

Vowels Under Stress Same Vowels Unstresseda. ªroguf ‘of horn’ ruªgat ‘horned’b. ªonzi ‘that’ (masc.) uªnazi ‘that’ (fem.)c. ªselu ‘village’ siªla ‘villages’d. ªrabut� ‘work’ r ��ªbotnik ‘worker’e. ªgrat ‘city’ gr ��ªdets ‘town’

The stepwise nature of Bulgarian vowel reduction has made it difficult to treat as a

unified phenomenon in classical generative phonology. In particular, the reduction process

cannot be analyzed as the elimination of non-high vowels, since the reduction of unstressed /a/

produces the non-high vowel [�]. However, under the Prominence Reduction approach, this

conundrum is easily solved: /e,o,a/ are defined as a group not according to distinctive features,

27

but according to sonority. Under the same logic, the vowels /i,u,�/ group together as a class of

low-sonority vowels. Thus, Bulgarian vowel reduction can be formally modeled using the

*Unstressed/X constraint family, along with faithfulness constraints for the features [round],

[front], and [high]. These constraints are ranked as follows in Bulgarian:

(22) Bulgarian Constraint Ranking

Max[round] Max[-high]Max[+front]

*Unstressed/a » *Unstressed/e,o»

*Unstresssed/i,u »*Unstressed/�

The Faithfulness constraints Max[round] and Max[+front] are undominated in this grammar,

indicating that underlying color specifications are always maintained in Bulgarian. The

constraint *Unstressed/a is also undominated, indicating that unstressed [a] will never occur in

Bulgarian. *Unstressed/e,o is dominated only by *Unstressed/a, so unstressed [e] and [o] will

likewise never occur. Finally, both *Unstressed/a and *Unstressed/e,o dominate the faithfulness

constraint Max[-high]. This domination allows the underlying [-high] specifications of /e,o/ to

be cast off and replaced with [+high] in the neutralizations /e,o/ > [i,u]. It also allows the

underlying [-high] specification of /a/ to be deleted without replacement in the neutralization /a/

> [�] (as noted earlier, schwa is represented as a targetless vowel, lacking any feature

specifications). This is illustrated in (23), demonstrating reduction of unstressed /e,o,a/. Note

that we can assume that all Dep[F] constraints are crucially dominated in Bulgarian, and will

therefore not be included in the tableaux for sake of brevity.

28

(23) Reduction of /e,o,a/ in Bulgarian

/roªgat/‘horned’

Max[round]

Max[front]

*Unstr-a

*Unstr-e,o

*Unstr-i,u

*Unstr�

Max[-high]

) [ruªgat] * *

[roªgat] *!

[r�ªgat] *! * *

[raªgat] *! *

/seªla/‘villages’

Max[round]

Max[front]

*Unstr-a

*Unstr-e,o

*Unstr-i,u

*Unstr�

Max[-high]

) [siªla] * *

[seªla] *!

[s�ªla] *! * *

[saªla] *! * * *

/graªdets/'town'

Max[round]

Max[front]

*Unstr.-a

*Unstr.-e,o

*Unstr.-i,u

*Unstr�

Max[-high]

) [gr�ªdets] * *

[gruªdets] *! *

[groªdets] *!

[graªdets] *!

Furthermore, the undominated position of faithfulness constraints for [round] and [front] also

prevent the reduction of unstressed /i,u/ to [�]:

(24) Nonreduction of /i,u/

/imeªna/ ‘names’(cf. ªime ‘name’)

Maxround

Max+front

*Unstr-a

*Unstr-e,o

*Unstr-i,u

*Unstr�

Max-high

) [imiªna] *

[emiªna] *!

[�miªna] *! *

/bukªvar/ ‘primer’

(cf. ªbukv� ‘letter’)

Maxround

Max+front

*Unstr-a

*Unstr-e,o

*Unstr-i,u

*Unstr�

Max-high

) [bukªvar] *

[bokªvar] *!

[b�kªvar] *! *

29

This account of Bulgarian vowel reduction not only accounts for the stepwise character

of the neutralizations, but it also corresponds well with instrumental investigation of Bulgarian

vowel reduction by Pettersson and Wood (1987a, 1987b). Their experiments suggest that the

vowels targetted by Bulgarian vowel reduction are dispreferred precisely because of their

prominence, not because of their featural content. Pettersson and Wood started their

investigation by first verifying the existence of acoustically neutralizing vowel reduction using

spectrographic evidence—the formant frequencies measured for unstressed /e,o,a/ were found to

coincide with those measured for the vowels /i,u,�/, respectively. Since there was no acoustic

difference in vowel quality reflecting these underlying vowel contrasts, we can state that

Bulgarian vowel reduction is acoustically neutralizing. However, using X-ray evidence, they

conclude that Bulgarian vowel reduction is not completely neutralizing in terms of articulation.

That is, they claim based on the X-ray evidence that unstressed /e/, /o/ and /a/ in Bulgarian

maintain the same (non-contrastive) tongue postures that are characteristic of their (respective)

pronunciation when stressed, but take on a high jaw position when unstressed—i.e., a height

similar to that seen for the vowels /i/, /u/, and /�/. For example, they describe the tongue

postures used for Bulgarian stressed /i/ as being more bunched and tense, while the tongue

postures used for Bulgarian stressed /e/ are more flat and lax—a distinction that was preserved in

unstressed syllables, despite the fact that they all had been acoustically neutralized to the quality

[i]. On the other hand, they describe the vowels /e,o,a/ as having on average 4 mm more

mandible depression when stressed than did the stressed vowels /i,u,�/. In other words, the

change from an underlying e-quality to a surface i-quality was not brought about by actively

changing the configuration or posture of the tongue, but by passively raising it by decreasing the

amount of jaw opening, thus creating a high front vowel. Similar changes in jaw position

30

account for the changes /o/ > [u] and /a/ > [�]. The important observation is that the articulation

of underlying /e,o,a/ was changed only enough to produce the vowels with the correct acoustic

quality. In this case, a partial articulatory neutralization is sufficient to cause a complete acoustic

merger. This shows that it is not necessarily the featural content of /e,o,a/ that is disfavored in

unstressed position, but their sonority. That is, a constraint like *Unstressed/a does not make

reference to the articulatory or featural qualities of [a], but to its prominence (i.e., duration and

amplitude). Therefore, this constraint will be equally non-violated by any vowel that is

acoustically [i], regardless of the articulations used to realize that quality. With this in mind, it

may be the case that Bulgarian unstressed /e,o/ are, in fact, [-high] on the surface, since they do

not adopt a high tongue posture. In this case, it would be necessary to reformulate the analysis

for Bulgarian vowel reduction already provided so as to replace articulatory features like [high]

with acoustic features like [high F2], or as suggested by Flemming (1995, this volume), to

include both articulatory and acoustic features. For example, the crucially dominated constraint

in the analysis already provided may not be Max[-high], but Max[-high F2]; whereas faithfulness

constraints for tongue body position remain undominated.

This type of Prominence Reduction analysis is easily extendable to similar phenomena in

other languages. One such example, Sri Lankan Portuguese Creole, is discussed in the next

section.

3.5 Prominence Reduction in Sri Lankan Portuguese:

Another interesting case of vowel reduction via prominence-reduction comes from Sri

Lankan Portuguese Creole (Smith, 1978). This language has seven phonemic vowels:

/i,u,e,o,4,a,c/. In this language, all vowels under stress are long. When stress shifts to a

different syllable, the vowel shortens. For example, the long stressed mid vowels in [ªoÛj] ‘eye’

31

and [ªt6eÛru] ‘fragrance’ correspond to short mid vowels in [oªjaÛ] ‘to see’ and [t6eªraÛ] ‘to smell

pleasant’. However, if the stressed vowel is low, it not only shortens, but also raises to the

corresponding mid vowel, as shown in the data below. This presents a case where low and mid

vowels are contrastive under stress, but are neutralized (in favor of the mid vowels) when not

under stress.

(25) Prominence Reduction in Sri Lankan Portuguese Creole

Low Vowels Under Stress Same Vowels Unstressedunstressed c > o ªccÛÛbr� ‘profession’ obªreÛru ‘manual worker’

ªnccÛÛmi ‘name’ nomiªnaÛ ‘nominate’unstressed 4 > e ªp44ÛÛd�r� ‘stone’ pedriªyaÛdu ‘ornamented with

stones’ªf44ÛÛru ‘iron’ f eªreru ‘blacksmith’

unstressed a > � ªbaÛÛjlu ‘dance’ b��jl ªdoÛr ‘dancer’ªbaÛÛrv� ‘beard’ b��rªveÛru ‘barber’

This vowel reduction pattern is quite similar to the Bulgarian pattern in that high-sonority vowels

are eliminated, and in that this is accomplished in a manner that respects underlying color

specifications.4 The difference between Sri Lankan Portuguese Creole and Bulgarian is that in

Bulgarian, the mid vowels /e,o/ are subject to this type of reduction, while in Sri Lankan

Portuguese Creole they are immune. To account for the Sri Lankan Portuguese Creole pattern of

vowel reduction, the same constraints seen in the Bulgarian analysis can be used, but the ranking

must be slightly different. As indicated below, the constraint *Unstressed/e,o is promoted above

Max[-high], although Max[-high] remains dominated by *Unstressed/a. Furthermore, it should

be pointed out that a constraint like *Unstressed/a does not refer only to specific quality named

32

in the constraint ([a]), but to any other vowel qualities that share its position in the sonority

hierarchy.1 Therefore, [4] and [c] also fall under the purview of this constraint.

(26) Sri Lankan Portuguese Creole Constraint Ranking

Max[round]Max[+front] Max[-high] *Unstressed/e,o »

*Unstressed/a» »

*Unstresssed/i,u »*Unstressed/�

The ranking of *Unstressed/a above Max[-high] ensures that the low vowel /a/ can lose its

underlying [-high] specification to emerge as the featureless vowel [�]. However, the ranking of

Max[-high] above *Unstressed/e,o ensures that unstressed /e,o/ will not reduce to [i,u]. (Further,

the undominated position of Max[round] and Max[+front] ensures that none of the vowels

/e,o,i,u/ will reduce to [�].) These rankings are illustrated in the following tableaux:

(27) Tableaux for Sri Lankan Portuguese Creole

/ncmiªna/'nominate'

Max[round]

Max[front]

*Unstr-a

Max[-high]

*Unstr-e,o

*Unstr-i,u

*Unstr�

) [nomiªnaÛ] *

[numiªnaÛ] *! *

[ncmiªnaÛ] *!

[n�miªnaÛ] *! * *

/bajlªdor/'dancer'

Max[round]

Max[front]

*Unstr.-a

Max[-high]

*Unstr.-e,o

*Unstr.-i,u

*Unstr�

) [b�jl ªdoÛr] * *

[bajlªdoÛr] *!

1 This being the case, it might be more appropriate to call the constraint *Unstressed/[low]. However, this wouldimply the parallel *Unstressed/[mid]. Since [�] is often (though not uncontroversially) classed as a mid vowel, thisnomenclature would cause confusion—as laid out in the preceding sections, [�] is an extremely low sonority vowelwhile “full” mid vowels like [e,o] are of medium sonority.

33

/oªja/'to see'

Max[round]

Max[front]

*Unstr.-a

Max[-high]

*Unstr.-e,o

*Unstr.-i,u

*Unstr�

) [oªjaÛ] *

[uªjaÛ] *! *

[�ªjaÛ] *! * *

3.6 Prominence Reduction and Undershoot

As described above, prominence-reducing vowel reduction is a phenomenon in which

relatively long vowels are replaced in unstressed position with shorter vowels of a similar

quality. As such, prominence-reducing vowel reduction is somewhat similar, at least

superficially, to the phonetic phenomenon of vowel undershoot. Undershoot refers to a situation

in which a given speech sound is articulated in a manner that does not fully instantiate the

canonical realization of that sound. This usually occurs in contexts where articulation time is

short; although the appropriate types of gestures are made (tongue fronting, jaw depression, etc.),

the magnitude of the gestures is not adequate to produce a canonical realization of the given

phone. When undershoot applies to vowels, the result is often referred to as "vowel reduction,"

as in the famous 1963 article by Bjorn Lindblom, "Spectrographic Analysis of Vowel

Reduction." Here, the term "vowel reduction" does not refer to the phonemic neutralization of

unstressed vowels, but to gradient changes in vowel articulation that can strike any vowel,

stressed or unstressed, under the appropriate conditions, such as under a fast speech tempo

(Lindblom 1963, Koopmans-van Beinum 1980). Impressionistically, vowel undershoot has been

described as vowel centralization: an undershot vowel might naively be described as a more [�]-

like realization of that vowel quality. However, as first illustrated by Lindblom (1963), this is

not an accurate description of vowel undershoot. Lindblom demonstrates that vowel undershoot

cannot be straightforwardly described as centralization. Instead, he suggests that undershoot

results from the displacement of a given vowel in the direction of the surrounding consonantal

34

environment. In many cases, this sort of coarticulatory effect will, in fact, equate with vowel

centralization. In other cases, it equates to vowel fronting, backing, raising, etc., dependent on

the articulations used in the surrounding consonantal environment. This hypothesis was further

investigated in Moon and Lindblom (1994), where changes in vowel duration were seen to effect

not only overall vowel quality (shorter vowels = less canonical), but also rate of formant change

going into and coming out of the vowel (shorter vowel = more rapid transitions). To sum up, a

given vowel phoneme will have a canonical target quality, a quality which may only be

instantiated in contexts where the vowel is rather long. In contexts where the vowel is shorter,

the articulators must move more quickly to produce the same result in less time. In certain

contexts, articulatory effort might not be increased by the amount needed to maintain the

canonical target quality, producing a surface quality that is less canonical and colored to some

degree by coarticulation with surrounding segments.

As such, vowel undershoot and prominence-reducing vowel reduction are somewhat

similar in terms of end result: Both phenomena result in vowels with shorter articulation times in

contexts of low duration such as in unstressed syllables or under fast speech tempos. Some have

claimed that prominence-reducing vowel neutralizations like those analyzed here are in fact the

phonological analogs of undershoot, and that the motivation for prominence-reducing vowel

reduction is the effort avoidance (Flemming, this volume; Kirchner, this volume). I take the

contrary position that these two phenomena are in fact phonetic and phonological inverses of

each other. That is, in vowel undershoot, decreased articulation time leads to a change in vowel

quality, and can be traced to a pressure to avoid effortful articulations (i.e., ones in which

articulator movement must be fast). In contrast, in prominence-reducing vowel reduction, a

change in vowel quality leads to a decrease in articulation time. This phenomenon, as presented

above, is not motivated by effort avoidance, but by the desire to unite phonological entities with

35

similar phonetic characteristics (enhancement). Although this distinction may at first glance

seem overly nice, it makes certain typological predictions that seem to be empirically supported.

The first such prediction concerns the nature of other, similar prominence reduction

phenomena. The enhancement motivation advanced here for prominence-reducing vowel

reductions predicts that it is prominence, not articulatory effort, that should be avoided in non-

prominent prosodic positions. For example, phonemically bimoraic vowels are longer than their

monomoraic counterparts, but not necessarily more effortful. However, many languages shorten

bimoraic vowels in unaccented positions (cf. Crosswhite 2000 for one such case). Since, ceteris

paribus, the gestures used for the monomoraic and bimoraic versions of a given vowel quality

are the same, it is hard to see how simply holding the vowel for a longer time make it inherently

more effortful. In fact, it would naively seem that the monomoraic vowel would be more

effortful, since the same articulatory gestures must be compressed into a shorter time period,

requiring faster articulator movement.

The second prediction concerns the types of phenomena that we would expect to see in

prominent positions. Under the effort-avoidance approach, we do not necessarily expect to see

any special phenomena affecting prominent positions. Instead, we would expect to see only

faithful realization of underlying qualities, even those underlying qualities that are the most

effortful. Instead, we often see prominent positions targeted for augmentation phenomena,

which is expected under the enhancement motivation for prominence alignment. For example,

many languages will lengthen underlyingly monomoraic vowels in stressed position—even when

this eliminates a phonological contrast, namely that of length. Similarly, some languages

increase the vocalic prominence of stressed vowels through vowel lowering (two examples of

this are discussed in the next section), despite the resulting loss of contrast.

36

In summary, then, although both prominence-reducing vowel reduction and vowel

undershoot produce similar end results, they seem to have different motivations. In particular,

the enhancement-based motivation for prominence-reduction more easily extends to cases of

prominence-reduction that cannot be analyzed as effort avoidance, as well as to cases of

augmentation in prominent prosodic positions.

Having said this, it should be pointed out that effort-avoidance is not irrelevant for the

analysis of vowel reduction: I merely claim that effort avoidance is not the motivating power

behind it. Effort avoidance does enter into the picture, however, when considering logically

possible alternatives to phonemic vowel reduction. According to Robert Kirchner (personal

communication), a vowel of any quality can be made to have extemely low sonority by making it

extremely short. If this is so, it remains to be explained why prominence reduction constraints

such as *Unstressed/e do not reduce a targetted vowel’s sonority simply by making it extremely

short while maintaining its quality. Doing so would require either very rapid articulator

movement in unstressed syllables or a tolerance for non-canonical, undershot vowel realizations.

Lack of phonemic reduction via use of rapid articulator movement could be seen in grammars

where effort avoidance constraints are low-ranked, whereas lack of phonemic reduction via

tolerance of undershoot could be seen in grammars where constraints on accuracy of articulation

are low-ranked. Phonemic vowel reduction of the type considered in this chapter will result

when both types of phonetic implementation constraint have a relatively high ranking. For an

example of the formal implementation of constraints on effort avoidance to account for gradient

sound changes, see Kirchner (this volume).

3.7 Other Examples of Prominence Reduction

In the examples of prominence-reducing vowel reduction discussed above, vowel

faithfulness constraints acted in a fairly straightforward manner to determine which subset of

37

vowels underwent reduction, as well as the types of neutralizations used. However, vowel

faithfulness constraints can interact with the *Unstressed/X constraint family in other, more

complex ways, resulting in additional sub-types of the Prominence Reduction phenomenon.

Reduction to [�]: For example, if the entire *Unstressed/X constraint family outranks

vowel faithfulness constraints (particularly those demanding preservation of color features), all

unstressed vowels will surface as the extremely low-sonority vowel [�]. This form of reduction

is common, for example, in English and in some non-standard varieties of Russian (Dedova

1988).

If, however, the *Unstressed/X constraint family outranks all but one faithfulness

constraint, the resulting pattern of reduction is one in which most unstressed vowels reduce to

[�], leaving one or two ‘survivors’ to surface without reduction. This type of pattern is observed,

for example, in some dialects of Italian, in which all vowels but /a/ undergo reduction to [�] in

pretonic position (Maiden 1995).

And finally, if all of the vowel faithfulness constraints outrank the entire *Unstressed/x

constraint family, no reduction will occur. But if just one of the constraints is demoted to the

bottom of the hierarchy, a pattern will result in which most unstressed vowels remain unreduced,

but one or two vowels are singled out for reduction to [�]. This type of pattern is observed in the

Sadzhava dialect of Ukrainian (Popova 1972), where /e,(/ reduce to [�] while all other vowels

remain unreduced, as well as in the Pavlikianski and Shirokolushki dialects of Bulgarian, where

only /i/ and only /i,e/ (respectively) undergo reduction to [�] (Stojkov 1968). Another well-

known example is the reduction of unstressed monomoraic /i,u/ to [�] in several dialects of

Slovene (Bezlaj 1939, ToporiÎi� 1976, Len�ek 1982).

38

Reduction Via Tensing: As just mentioned, if all vowel faithfulness constraints outrank

the entire *Unstressed/X constraint family, they cannot cause any vowel reduction. However, if

faithfulness constraints on [ATR] are demoted to a position below *Unstressed/(,o, a form of

reduction will occur in which the unstressed mid lax vowels /(,o/ undergo a minimal form of

raising, surfacing as the slightly less sonorous vowels [e,o]. This form of reduction is actually

quite common, occurring most notably in dialects of Italian and Catalan.

Since this is the only form of prominence-reducing vowel reduction that completely

respects underlying height and color features, it is predicted to be quite common. In earlier work

(Crosswhite 2001), I have computed the factorial typology of the constraint set proposed here. It

predicts that reduction-via-tensing should be able to co-occur with many other patterns of vowel

reduction. This type of pattern is in fact attested in languages like Trigrad Bulgarian (Stojkov

1963), where a reduction-via-lowering pattern (/o,o/ > [a], a case of contrast enhancing

reduction, to be discussed shortly) is accompanied by reduction-via-tensing (/(/ > [e]). A similar

pattern is found in Majorcan Catalan (Recasens 1991).

It is interesting to note at this juncture that the opposite type of pattern, reduction-via-

laxing, has also been reported to occur in two different languages: Standard Slovene and

northeastern dialects of Brazilian Portuguese. However, instrumental analysis of the Slovene

case casts doubt on this interpretation of the facts. This type of reduction is discussed as

reduction via feature deletion in section 2.5.

3.8 Prominence Reduction in Additional Contexts

Finally, it should be recalled that Prominence Reduction constraints are based on the

combination of two prominence scales—all of the Prominence Reduction constraints examined

in this section deal with (1) an accentual prominence scale, and (2) a vocalic prominence scale

39

(vowel sonority). However, the parameter accented vs. unaccented is not the only type of

prominence that can interact with vowel quality in the fashion under consideration. For example,

in addition to accentual prominence, there is also weight-based prominence. And indeed, there

are cases of vocalic neutralizations that appear to be motivated by this sort of weight-based

Prominence Reduction constraint. One such case, Carniolan Slovene, will be discussed now in

order to show the Prominence Reduction mechanism can be extended to this type of case.

Many dialects of Slovene exhibit vowel reductions. Although the exact nature of these

reductions vary from dialect to dialect, one factor that is shared by most of the Slovene reduction

patterns is the conditioning environment: accented bimoraic vowels are immune to reduction,

while both unaccented vowels and accented monomoraic vowels are subject to neutralizations.

Additionally, surface bimoraic vowels are only allowed in accented syllables in Slovene; if

accent moves off an underlyingly long vowel, it concomitantly shortens: cf. [ªpl((ÛÛma] ‘tribe’ v.

[pl((ªmeÛna] ‘tribes’. Thus it is appropriate to characterize Slovene vowel reduction as targeting

monomoraic vowels, whether accented or not. For example, in many Carniolan dialects of

Slovene (upon which the literary language is based), monomoraic /i,u/ reduce to [�]. As

mentioned in section 0 above, reduction of /i,u/ to [�] while all other vowel qualities remain

unreduced is a special case of reduction to [�] in which most, but not all, vowel faithfulness

constraints outrank the entire family of relevant Prominence Reduction constraints. In this case,

the relevant Prominence Reduction family is not *Unstressed/X, but *Monomoraic/X. That is,

just as *Unstressed/X family requires that non-prominent unstressed positions be filled with low-

sonority vowels, the constraint family *Monomoraic/X requires that all monomoraic vowels be

low in sonority (while the more prominent bimoraic vowels are left unaffected).5 The

*Monomoraic/X family is illustrated below:

40

(28) *Monomoraic/a » *Monomoraic/e,o » *Monomoraic/i,u » *Monomoraic/�

For example, the constraint *Monomoraic/a says that the segment [a] is too sonorous to be only

monomoraic: to escape violation, all underlying /a/ must either lengthen, or undergo a reduction

in sonority. That is, if left unbridled, this constraint family would reduce all monomoraic vowels

to [�]. However, in Slovene, this constraint family is not left unbridled: High-ranking vowel

faithfulness constraints block reduction to [�] for most of the underlying vowel qualities of

Slovene:

(29) Constraint ranking for Slovene

Max[+low],Max[-high]

»

*Monomoraic/a »*Monomoraic/e,o »*Monomoraic/i,u »

*Monomoraic/�

»Max[+high],Max[round],Max[+front]

By ranking Max[+low] and Max[-high] above all the monomoraic vowel reduction constraints,

the nonhigh vowels are completely immune to reduction: any decrease in sonority of a

monomoraic mid or low vowel would require violating one of these two undominated

constraints. However, the high vowels are not protected in this manner: The ranking of

*Monomoraic/� above faithfulness constraints for color and [+high] leaves the high vowels open

to reduction: This analysis correctly predicts that bimoraic /i,u/ are not subject to this form of

reduction.

3.9 Comparison of Reduction Types

At this point, I have proposed that vowel reduction is not a single phenomenon, but two

independent, formally distinct phenomena: contrast-enhancement, and prominence-reduction. It

is worth noting at this time that these two reduction phenomena are sometimes indistinguishable.

For example, under both types of reduction, unstressed [e,o] are disfavored: either because they

41

are noncorner (contrast-enhancement), or because they are moderately sonorous (prominence-

reduction). Furthermore, given the Optimality-Theoretic formulation of these two phenomena, it

is entirely possible for either of these categories to give rise to non-canonical vowel reduction

patterns which only partially instantiate the ideas of prominence-reduction or contrast-

enhancement. For example, high-ranking for a faithfulness constraint might block a certain

vowel from undergoing reduction (e.g., high rank for Max[+low] could block reduction of

unstressed /a/). With this in mind, any case where mid vowels are neutralized via raising to [i,u]

is ambiguous: are they being raised to produce corner vowel qualities, or to produce low-

sonority vowel qualities? In some cases, it is possible to distinguish the two. For example, if

unstressed /a/ also reduces, for example, to [�], it is a case of prominence-reduction: the

reduction /a/ > [�] decreases sonority, but does not produce a corner vowel. Similarly, if any

vowel reduces to [a], it is a case of contrast-enhancement. For example, if /e/ > [i] but /o/ > [a]

(as in Russian), we have a case of contrast enhancement since one of these reduction, /o/ > [a]

produces a corner vowel, but does not reduce sonority. However, in cases where both /e,o/

reduce to [i,u], and unstressed /a/ remains unreduced, no definitive categorization can be made:

it could be that /a/ is immune because Lic-Noncorner/Stress does not affect unstressed /a/, or it

could be the case that high-rank of Max[+low] blocks reduction of /a/.

4. Two Pattern Vowel Reduction Systems:

Finally, it appears that more than one type of vowel reduction can occur in the same

language. In this type of vowel-reduction language, there are two sets of neutralizations that

occur in unstressed syllables. One, typically a “moderate” form of reduction, takes place in

certain unstressed syllables, while a more “extreme” form of reduction takes place in the

remaining syllables. This is the case, for example, in most dialects of Russian. Here, unstressed

/o/ and /a/ neutralize. In the syllable immediately preceding the stress, this generates the surface

42

vowel quality [a], while in other unstressed syllables it generates the surface quality [�]. This is

demonstrated below with data from standard Russian:

(30) Russian Two-Pattern Vowel Reduction

in stressed σ in immediately in other pre-stress σ unstressed σ gloss

ªdom (nom. sg.) daªma (nom. pl.) d��maªvoj (adj.) ‘house’ªgol�vu (acc.) gaªlofka (diminutive) g��laªva (nom. sg.) ‘head’ªkam-in- (nom. sg.) kam-ªn-ej (gen. pl.) k��m-iªn-ist�j (adj.) ‘stone’ªdal-iji (comp.) daªl-ok-ij (adj.) d��l-iªko (adverb) ‘far’

A similar pattern is seen in some southern Russian dialects, where unstressed /e/ and /o/ both

neutralize to [a] in the syllable immediately preceding the stress, but reduce either to [i] (for

underlying /e/ and for underlying /o,a/ preceded by a palatalized consonant) or to [�] (underlying

/o,a/ elsewhere). Other such patterns are found in Rhodope Bulgarian dialects (Miletich 1936),

certain Italian dialects (Maiden 1995), and standard Brazilian Portuguese (Dukes 1993,

Redenbarger 1981), to name a few. These patterns are briefly described in the table below:

43

(31) Examples of Two-Pattern Vowel Reduction

Language “Moderate” Reduction “Extreme” Reductionsouthern Russian Unstressed /e,o/ both

neutralize to [a] in the syllableimmediately preceding thestress.

In all remaining unstressedsyllables, /o,a/ reduce to [�] (or[i] following a palatalizedconsonant) and unstressed /e/reduces to [i].

ContemporaryStandard Russian

Unstressed /o/ neutralizes to[a] in the syllable immediatelypreceding the stress

Unstressed /o/ and /a/neutralize to [�] in theremaining unstressed syllables.

RhodopeBulgarian

Unstressed /e/ and /o/ bothneutralize to [a] in any syllablepreceding the stress

Unstressed /e/ and /o/neutralize to [i] and [u],respectively in post-tonicpositions. In the samecontexts, unstressed /a/neutralizes to [�].

dialectal Italian All unstressed vowels except/a/ neutralize to [�] in anysyllable preceding the stress.

All unstressed vowelsneutralize to [�] post-tonically.

Standard BrazilianPortuguese

Unstressed /(,o/ neutralize to[e,o], respectively, inunstressed non-word-final andnon-word-initial syllables.

Unstressed /(,e/ and /o,o/neutralize to [i] and [u],respectively word-final andword-initial unstressedsyllables. Also, unstressed /a/becomes [�].

The “extreme” vowel reduction processes share some common features. First, the type of

neutralization seen in “extreme” reduction is always sonority-decreasing—in contrast with the

“moderate” reductions, which can be sonority-increasing (cf. the Russian and Bulgarian change

of unstressed /o/ to [a]). Second, they seem to target those unstressed syllables that are the most

durationally impoverished. This fact makes it possible to identify “extreme” vowel reduction as

a case of neutralization caused by prominence-reduction targeting durationally-impoverished

unstressed syllables, while the changes seen in “moderate” reduction can be ascribed to a second

reduction phenomenon that targets unstressed syllables in general (usually contrast-enhancing

reduction, but cf. the dialectal Italian pattern). An interesting parallel that can be drawn at this

44

point is that contrast-enhancing reduction can occur in either stress-timed or syllable-timed

languages, but prominence-reduction appears to occur only in stress-timed languages (or

dialects). Based on evidence from various languages with vocalic prominence-reduction

phenomena, I conclude that “extreme” reduction (a form of prominence reduction found in two-

pattern systems) occurs in durationally impoverished syllables. In Crosswhite (1999), I argue

that such durationally impoverished syllables are in fact nonmoraic.

As noted above, the contexts for “extreme” vowel reduction comprise the most

durationally impoverished syllables found in a given language. Put another way, the “moderate”

vowel reductions are only found in those unstressed syllables which have slightly greater

duration than the other unstressed syllables. Two examples—standard Russian and Brazilian

Portuguese—are discussed in the paragraphs below.

The immediately pretonic syllable in Russian has long been recognized as having a

special durational status—it is much longer than other unstressed syllables, and can sometimes

even be longer than the stressed syllable (this is often the case, for example, with words where

the stressed vowel has a low inherent duration ([i,u]), and the immediately-pretonic vowel has a

high inherent duration ([a]).) This difference in the duration of Russian unstressed syllables is

significant enough to have been accurately noted by ear and described by 19th-century Russian

grammarians. This phenomenon is also easily observed indirectly when listening to Russian

speech: at conversational speech tempos, non-immediately pretonic unstressed vowels are often

completely or near-completely elided, but immediately pretonic ones are immune to this process.

For example, the word /xoroª6o/ ‘good’ has the citation pronunciation [x�raª6o], however is often

pronounced [xraª6o], but never *[x�rª6o].

45

The special durational status of Russian immediately pretonic syllables has also been

confirmed experimentally a number of times (see, for example, Zlatoustova 1981). Investigators

have also noted a link between this type of durational effect and the presence vs. absence of

varying degrees of vowel reduction in Russian dialects (Vysotskii 1973, Al’mukhamedova &

Kul’sharipova 1980, Kasatkina et al. 1996, etc.). For example, in a certain group of north-central

Russian dialects (the Vladimir-Volga Basin group), the immediately pretonic syllable has the

same special durational status that is seen in Contemporary Standard Russian—similarly, in

those dialects the immediately pretonic syllable also has a special status with respect to vowel

reduction, in that all other unstressed syllables are subject to vowel reduction, but the

immediately pretonic one is immune (that is, unstressed /o/ remains [o]). In another case, the

immediately pretonic syllable in certain southern Russian dialects predictably displays either

“moderate” reduction or “extreme” reduction. As shown by Kasatkina et al. (1996), this

predictable variation in vowel reduction pattern is accompanied by changes in prosody—when

the immediately pretonic syllable displays “moderate” reduction, this syllable has relatively

longer duration; when it displays “extreme” reduction, it has a duration about equivalent to that

of the other unstressed syllables.

A similar situation is seen in standard Brazilian Portuguese. For example, in instrumental

work by Major (1992), it was found that post-tonic syllables in Brazilian Portuguese undergo

greater shortening than do pretonic syllables. Based on this and other evidence, Major

hypothesizes that the post-tonic syllables in Brazilian Portuguese are stress-timed, while the

pretonic syllables are syllable-timed. I will assume here that this is equivalent to saying that

post-tonic syllables in Brazilian Portuguese can be nonmoraic, while pre-tonic ones cannot.

Major also suggests that this dichotomy in the durational properties of pre- and post-tonic

unstressed syllables is related to the two different vowel reduction patterns seen in this language,

46

although in a way that is slightly different from the relationship hypothesized here—Major

hypothesizes that the sonority-decreasing vowel neutralizations seen post-tonically ((,e>i;

o,o>u; a>�) exist in order to heighten the effect of the post-tonic shortening. Since Major also

observed some shortening pretonically (i.e., in faster speech tempos), he suggests that Brazilian

Portuguese is in the process of converting from syllable-timing (which is typical of several

Romance languages, including Spanish and Italian) to stress-timing. Judging from the comments

made by Brakel (1985) and de Carvalho (1988-92), this process is already at a more advanced

stage in the European (Iberian) variant of Portuguese. Both researchers notes that one of the

differences between Brazilian and European Portuguese is the absence in the European variant of

the “moderate” vowel reductions seen in the Brazilian variant. That is, in European Portuguese,

all unstressed syllables are subject to “extreme” reduction.6 They further note that European

Portuguese differs from Brazilian Portuguese in the type of manipulation that an unstressed

vowel endures—in the European variant, unstressed vowels (including pretonic vowels) are

subject to extreme shortening, which often results in the devoicing and/or complete deletion of

the vowel—a type of pronunciation that is less typical for Brazilian Portuguese.

My general approach to accounting for these observations is that stress-timed

languages—with their ultra-short vowels that are prone to devoicing or deletion—are languages

which allow some subset of their vowels to surface without associated moras. The exact

distribution of these nonmoraic vowels varies from language to language—in Brazilian

Portuguese, they are post-tonic; in Russian they cannot occur immediately before the stressed

syllable; and in Iberian Portuguese they seem to occur in most unstressed syllables.

5. Conclusion

In this chapter, I have considered numerous cases of vowel reduction, all of which fall

into one of two broad groups: reduction based on the concept of prominence, and reduction

47

based on contrast-enhancement. The prominence-based category is motivated by the desire to

unite different elements with similar prominence characteristics, and the contrast-enhancing

category is motivated by the desire to avoid perceptually challenging vowel qualities in all but

the most perceptually-favorable positions. These two desiderata are encoded in an Optimality-

Theoretic analysis using, respectively, prominence reduction constraints, and licensing

constraints.

This dichotomy is especially useful in resolving “reduction paradoxes.” For example,

consider the reduction patterns of Bulgarian (/i,e,�,a,o,u/ reduces to [i,u,�]) and Belarusian

(/i,e,a,o,u/ reduces to [i,u,a]). If these two reduction patterns are superficially compared, we

might arrive at the anomalous conclusion that the vowel quality [a] is both highly marked (it

undergoes reduction in Bulgarian) and highly unmarked (it serves as a reduction vowel in

Belarusian). Another example of this sort of “reduction paradox” is observed when comparing

between Bulgarian and Sri Lankan Portuguese Creole (/4,a,c/ > [e,�,o])—the mid vowels seem

to be both marked (they undergo reduction in Bulgarian) and unmarked (they serve as reduction

vowels in Sri Lankan Portuguese Creole). The analysis for vowel reduction advanced here

resolves these paradoxes, by positing separate families of Markedness constraints, based on

distinct phonetic motivations. Specifically, the basic claim of this analysis is that vowels are not

“absolutely” marked, and, therefore, vowel reduction patterns cannot be used as indicators of

absolute vocalic markedness.

This appeal to two independent motivations for vowel reduction suggests that, in fact,

there is no such thing as a monolithic concept of markedness, at least as far as unstressed vowels

are concerned. In effect, the phonological concept of markedness has been replaced by phonetic

considerations, which are encoded in phonology using phonetically-motivated constraints.

48

References:

Al'mukhamedova, Z.M. and P.Ë. Kul'sharipova. (1980). Reduktsiia glasnikh i prosodiia slova vokaiushchikh russkikh govorakh. Kazan': Izdatel'stvo Kazan'skogo Universiteta.

Archangeli, Diana, and Douglas Pulleyblank. (1994). Grounded Phonology. Cambridge, MA:MIT Press.

Baroni, Marco. (1996). An acoustic study of Italian unstressed mid-vowels. Paper presented atAcoustical Society of America, Dec. 4 1996, Hawai’i.

Bezlaj, Fr. (1939). Oris slovenskega knjiÞnega izgovora. Ljubljana: Natisnila u�iteljskatiskarna.

Bidwell, Charles E. (1969). Outline of Slovenian Morphology. Pittsburgh: University Center forInternational Studies, University of Pittsburgh.

Bloomfield, Leonard. (1957). Eastern Ojibwa: Grammatical Sketch, Texts, and Word List.Ann Arbor: University of Michigan Press.

Brakel, Arthur. (1985). Reflections on the analysis of exceptions to the rule of IberianPortuguese vowel reduction. Hispanic Linguistics 2:1. 63-85.

Browman, Catherine P and Louis Goldstein. (1992). 'Targetless' Schwa: An ArticulatoryAnalysis. In Docherty and Ladd (eds.), Papers in Laboratory Phonology, II: Gesture,Segment, Prosody. Cambridge: Cambridge University Press. 26-67.

Crosswhite, Katherine. (1998). Segmental vs. Prosodic Correspondence in Chamorro.Phonology 15:3. 281-316.

Crosswhite, Katherine. (2001). Vowel Reduction in Optimality Theory. New York: Routledge.de Carvalho, Joaquim Brandao. (1988-92). Reduction vocalique, quantite et accentuation: pour

une explication structurale de la divergence entre portugais lusitanien et portugaisbresilien. Boletim de Filologia XXXII . 5-26.

Dedova, O. V. (1988). Nabl'udeniia nad vokalizmom sovremennogo moskovskogoprostorechiia. Vestnik Moskovskogo Universiteta. Series 9 "Filologiia," no. 2. 28-36.

Diehl, Randy. (1986). Remarks on Stevens' quantal theory of speech. JPh 17. 71-78.Dukes, Beatriz. (1993). Vowel Reduction and Underspecification in Brazilian Portuguese.

UCLA Occassional Papers in Lingusitics 13: Papers in Phonology. 21-48.Flemming, Edward. (1995). Auditory Representations in Phonology. Ph.D. dissertation,

UCLA.Fry, D. B. (1979). The Physics of Speech. Cambridge: Cambridge University Press.Gruzov, L.P. (1960). Sovremennyi mariiskii iazyk: fonetika. IOshkar-Ola: Mariiskoe knizhnoe

izd-vo.Gordon, Matthew. (this volume). Syllable Weight.Hualde, Jose. (1991). Basque Phonology. London: Routledge.Jean-Luc Schwartz, Louis-Jean Boe, Nathalie Vallee, and Christian Abry. (1997). The

Dispersion-Focalization Theory of vowel systems. JPh 25. p. 255-286.Jiang-King, Ping. (1996). An Optimality Account of Tone-Vowel Interaction in Northern Min.

Ph.D. dissertation, Chinese University of Hong Kong.Johnson, D.H. (1980). The relationship between spike rate and synchrony in responses of

auditory-nerve fibers to single tones. JASA 68. 1115-1122.Kamprath, Christine. (1991). Toward a Restrictive Model of Unstressed Vowel Reduction. Ms.

of 1991 LSA presentation.

49

Kasatkin, L.L. (1989). Russkaia Dialektologiia. Moscow: Prosveshcheniie.Kasatkina, R.F., et al. (1996). Osobennosti prosodii slova v iuzhnorusskikh govorakh. In T.

Nikolaeva (ed.), Prosodicheskii stroi russkogo iazyka. Moscow: RAN.Keating, Patricia. (1983). Comments on the jaw and syllable structure. JPh 11, 401-406.Kenstowicz, M. (1994). Sonority-Driven Stress. Ms., MIT.Koopmans-van Beinum, F.J. (1980). Vowel Contrast Reduction, an acoustic and perceptual

study of Dutch vowel in various speech conditions. Ph.D. dissertation, University ofAmsterdam.

Krivitskii, A.A. and A.I. Podluzhnyi. (1994). Uchebnik belorusskogo iazyka dliasamoobrazovaniia. Minsk: Vysh£ishaia Shkola.

Ladefoged, P., R. Harshman, L. Goldstein, and L. Rice. (1971). Vowel articulation and formantfrequencies. UCLA Working Papers in Phonetics 38. 16-40.

Lehiste, Ilse. (1961). The Phonemes of Slovene. International Journal of Slavic Linguisticsand Poetics IV . 48-66.

Lehiste, Ilse. (1970). Suprasegmentals. Cambridge, MA: MIT.Len�ek, Rado. (1982). The Structure and History of the Slovene Language. Columbus, OH:

Slavica.Lieberman, P. (1971). Uniquely human: The evolution of speech, thought, and selfless

behavior. Cambridge, MA: Harvard University Press.Liljencrants, J. and B. Lindblom. (1972). Numerical simulations of vowel quality systems: the

role of perceptual contrast. Lg 48. 839-862.Lindblom, B. (1963). Spectrographic Study of Vowel Reduction. JASA 35. 1773-81.Lindblom, B. (1983). Economy of Speech Gestures. In P. F. MacNeilage (ed.), The Production

of Speech. New York: Springer-Verlag. 217-246.Lindblom, B. (1986). Phonetic universals in vowel systems. In John Ohala and Jeri Jaeger

(eds), Experimental Phonology. Orlando: Academic Press. 13-44.Lindblom, B. and I. Maddieson. (1988). Phonetic universals in consonant systems. In L. M.

Hyman and C. N. Li (eds.), Language, Speech and Mind. London: Routledge. 62-78.Lutta, C.M. (1923). Der Dialekt von BergÕn und seine Stellung innerhalb der raetoromanischen

Mundarten Braubeundens. Supplement to Zeitschrift fÕr Romanische Philologie 71.Maddieson, Ian. (1984). Patterns of Sounds. Cambridge: Cambridge University Press.Maiden, Martin. (1995). Evidence from the Italian Dialects for the Internal Structure of Prosodic

Domains. In John Charles Smith and Martin Maiden (eds.), Linguistic Theory and theRomance Languages. Amsterdam: Benjamins. 115-31.

Major, Roy C. (1992). Stress and Rhythm in Brazilian Portuguese. In Dale A. Loike andDonald P. Macedo (eds.), Romance Linguistics: The Portuguese Context. Westport:Bergin and Garvey. 3-30.

Majors, Tivoli. (1998). Stress-dependent vowel harmony: phonological analysis and phoneticexplanation, Ph.D. dissertation, University of Texas, Austin.

Malsch, Derry, and Roseanne Fulcher. (1989). Categorizing Phonological Segments: TheInadequacy of the Sonority Hierarchy. In Robert Corrigan, Fred Eckman, and MichaelNoonan (eds.), Linguistic Categorization. Oxford: Clarendon. 69-80.

Miletich, L. (1936). ’Akane’ i ‘pulnoglasie’ v× tsentralniia rodopski govor×. Makedonski

pregled× X:1-2. 26-30.Miller, Patricia. (1972). Vowel Neutralization and Vowel Reduction. CLS 8. 482-489.

50

Moon, Seung-jae, and Bjorn Lindblom. (1994). Interaction between Duration, Context, andSpeaking Style in English Stressed Vowels. JASA 96. 40-55.

Munro, Pamela and Peter John Benson. (1973). Reduplication and Rule Ordering in Luiseño.IJAL XXXIX . 15-21.

Nathan, Geoffrey. (1989). Preliminaries to a Theory of Substance: The Substance of Sonority.In Robert Corrigan, Fred Eckman, and Michael Noonan (eds.), Linguistic Categorization.Oxford: Clarendon. 55-67.

Palková, Zdena. Fonetika a Fonologie Cestiny. Prague: Universita Karlova.Perkell, J. S. and W. L. Nelson. (1985). Variability in production of the vowels /i/ and /a/.

JASA 77. 1889-1895.Perkell, J.S. and M. H. Cohen. (1986). An indirect test of the quantal nature of speech in the

production of the vowels /i/, /a/, and /u/. JPh 17. 123-133.Perkell, J.S. and W. L. Nelson. (1982). Articulatory targets and speech motor control: A study

of vowel production. In S. Grillner, A. Persson, B. Lindblom, and J. Lubker (eds.),Speech Motor Control. Oxford: Pergamon. 187-204.

Perrone, Charles A. and Linda Ledford-Miller. (1985). Variation in Pretonic /e/ in BrazilianPortuguese: Preliminary Studies with Popular Music of the Northeast and Rio de Janeiro.Hispania 68:1. 154-159.

Peterson, Gordon, and Ilse Lehiste. (1960). Duration of syllable nuclei in English. JASA 32.693-703.

Pettersson, Thore, and Sidney Wood. (1987a). Vowel Reduction in Bulgarian and itsImplications for Theories of Vowel Production: A Review of the Problem. FoliaLinguistica XXI:2-4 . 261-279.

Pettersson, Thore, and Sidney Wood. (1987b). Vowel Reduction in Bulgarian: The PhoneticData and Model Experiments. Folia Linguistica XXII:3-4 . 239-262.

Pisoni, D. (1980). Variability of vowel formant frequencies and the quantal theory of speech: Afirst report. Phonetica 37. 285-305.

Popova, T.V. (1972). Paradigmaticheskie konsonantnye riady cheredovanii v iugo-zapadnykhukrainskikh dialektakh (na materiale govora s. Sadzhava). In G.P. Klepikova (ed.),Karpartskaia Dialektologiia i Onomastika. Moscow: Nauka. 179-239.

Prince, Alan and Paul Smolensky. (1993). Optimality Theory: constraint interaction ingenerative grammar. To appear from MIT Press. TR-2, Rutgers University CognitiveScience Center.

Recasens, Daniel. (1991). Fon£tica descriptiva del Catal�. Barcelona: Institut d’EstudisCatalans.

Redenbarger, Wayne J. (1981). Articulator features and Portuguese vowel height. HarvardStudies in Romance Languages 37. Cambridge, MA: Dept. of Romance Languages andLiteratures of Harvard University.

Scatton, E.A. 1984. A Reference Grammar of Modern Bulgarian. Columbus, OH: Slavica.Smith, Ian R. (1978). Sri Lanka Creole Portuguese Phonology. Dravidian Linguistics

Association. Vanchiyoor: Kerala University Cooperative Stores Press.Steriade, Donca. (1994a). Licensing by Cue. Ms., UCLA.Steriade, Donca. (1994b). Positional Neutralization and the Expression of Contrast. Ms,

UCLA.Stevens, Kenneth. (1986). On the quantal nature of speech. JPh 17. 3-45.

51

Stojkov, Stojko. (1963). Akane v govora na s. Trigrad, Devinsko. B×lgarski Ezik, XIII:1 . 8-21.

Stojkov, Stojko. (1968). B×lgarska dialektologiia. Sofia: Nauka i isskustvo.

ToporiÎi�, JoÞe. (1976). Slovenska slovnica. Maribor: ZaloÞba Obzorja.

Uba̧ czyk, Stanis¶aw. (1953). Zarys dialektologii Polskiej. Warszawa: PWN.Vaillant , Andre. (1958). Grammaire comparee des langues slaves, Vol. 1 Phonetique. Lyon:

IAC.Vysotskij, S.S. (1973). O zvukovoi strukture slova v russkikh govorakh. In Issledovaniia po

russkoi dialektologii. Moscow: Nauka.Zlatoustova, L. V. (1981). Foneticheskie edinitsy russkoi rechi. Moscow: Izdatel’stvo

Moskovskogo universiteta.Zoll, Cheryl. (1996). Parsing below the Segment in a Constraint-Based Framework. Ph.D.

dissertation, UC Berkeley.

1 But see Casali (1998) for an argument that the acoustically-grounded approach to the relationship between [ATR]and [low] offers a superior account of certain vowel harmony facts.2 It is also worth noting here that this approach does not necessarily predict that non-reduced mid central vowels,such as English [¥], should have the same position in the sonority hierarchy as [�]. Hence, Peterson and Lehiste's(1960) finding that English [�] is in fact slightly longer than [,�8,(] is not contrary to the position outlined above,since they classify the vowels in tuck and tug as [�].3 Note that in the pronunciation norm of Sofia and other western areas of Bulgaria, vowel reduction is weaker thanin the eastern areas, or even entirely absent. Since Sofia pronunciation defines the standard, many vowel-reducingspeakers attempt to suppress vowel neutralizations when speaking in formal registers. In particular, suppression ofthe reduction /e/>[i] is quite common, and lack of this suppression is rather stigmatized—cf. Scatton 1984.4 Again, it may be the case that the underlyingly low vowels in unstressed position maintain some articulatorydifferences, when compared with phonemically mid vowels. Lacking any concrete data on this point, I will continueto use traditional articulatory features in this and subsequent analyses.5 It is important to bear in mind that the monomoraic category is in fact intermediate in prominence: it is lessprominent than bimoraicity, but more prominent than nonmoraicity: µ µ prom> µ prom> ∅. For this reason, themonomoraic category can be the basis for both a Prominence Reduction constraint family as in the main text (whichI designate *Monomoraic/X), as well as a Prominence Alignment family (*µ/X). The former places limitations onwhich vowels are too sonorous to be monomoraic (as opposed to bimoraic), while the latter places limitations onwhich segments (usually consonants) are not sonorous enough to be moraic. For example, *Monomoraic/a says that[a] is too sonorous to be only monomoraic, while *µ/t says that [t] is not sonorous enough to contribute phonologicalweight. These two distinct families differ in their inherent rankings: *Monomoraic/a » *Monomoraic/t but *µ/t »*µ/a.6 Other differences between Brazilian and European Portuguese vowel reduction include: (1) in EuropeanPortuguese, vowel reduction is obligatory even in slow speech; in Brazilian Portuguese the pretonic “moderate”reduction is optional in careful (citation) speech; (2) in European Portuguese, unstressed /e/ reduces to [�], whereasin Brazilian Portuguese it reduces to [i] (post-tonically).